TVBS Debuts Self-Developed Real-Time AI Translation at NVIDIA Keynote

TVBS, a Taiwanese media organization, made headlines by deploying its self-developed real-time AI translation system during NVIDIA CEO Jensen Huang's keynote at GTC Taipei 2026 on June 1. The system delivered near-simultaneous Chinese subtitles that viewers initially mistook for official NVIDIA translations, showcasing a rare instance of a news organization independently developing live translation for a large-scale broadcast. Online viewers praised the translation quality, with comments including 'The quality of this real-time translation is amazing,' and 'The live speech recognition quality is surprisingly good.' Some viewers assumed the subtitles came from NVIDIA's official translation team. The achievement demonstrates how Taiwan's media industry is adapting to AI-driven transformation, with TVBS positioning itself as a leader in integrating proprietary AI development with real-world broadcast applications.

The company completed optimization and deployment of the system in just nine days — a process that traditionally requires months of dedicated training, according to Andy Yang, manager of TVBS AI Future Technology Department. Yang explained that the biggest challenge in real-time translation lies in balancing 'content accuracy' with 'reading fluency' under extremely tight time constraints. Huang's speaking style is highly improvisational and frequently switches between English, Mandarin and Taiwanese, while audience applause and cheering often occur simultaneously. Applying standard off-the-shelf translation software in such an environment would likely result in sentence segmentation errors and unclear semantic interpretation. Rather than pursuing 'zero-latency' translation at all costs, TVBS designed a subtitle presentation mechanism specifically tailored to the keynote format. Prior to the event, the TVBS team conducted extensive cross-platform testing on multiple AI translation models, evaluating latency and error rates. By introducing a carefully calibrated broadcast delay, the system gained additional time to process semantic meaning and contextual alignment, enabling smoother subtitles matching natural Chinese reading habits.

The team also built a dedicated knowledge base for NVIDIA GTC Taipei, compiling AI industry terminology, product names and technical keywords in advance, significantly improving translation quality and the accuracy of professional content delivery. TVBS's real-time translation system can be customized based on different broadcasting scenarios. Yang emphasized that the key to the rapid nine-day deployment was cross-department collaboration. Engineering teams handled hardware tuning and signal integration, while internal technical and content teams managed ongoing model training and refinement. Through continuous iteration and adjustment, the team incorporated Taiwanese linguistic nuances and natural Chinese phrasing into the system, enhancing fluency and readability beyond what generic AI models alone could achieve, ultimately shaping what Yang called a distinct 'TVBS AI style.' This approach reflects the company's '3T' philosophy — Truth, Trust and Technology. In his keynote at the Taipei Music Center, Huang focused on AI infrastructure, next-generation computing platforms and NVIDIA's latest technological roadmap. GTC Taipei 2026 attracted significant international attention as AI continues to reshape global industries. Through its own transformation journey, TVBS hopes to help lead Taiwan's media industry into a new era shaped by AI.